mirror of https://code.videolan.org/videolan/x264
Improve dct-a.S Performance by Using SVE/SVE2
Imporve the performance of NEON functions of aarch64/dct-a.S by using the SVE/SVE2 instruction set. Below, the specific functions are listed together with the improved performance results. Command executed: ./checkasm8 --bench=sub Testbed: Alibaba g8y instance based on Yitian 710 CPU Results: sub4x4_dct_c: 528 sub4x4_dct_neon: 322 sub4x4_dct_sve: 247 Command executed: ./checkasm8 --bench=sub Testbed: AWS Graviton3 Results: sub4x4_dct_c: 562 sub4x4_dct_neon: 376 sub4x4_dct_sve: 255 Command executed: ./checkasm8 --bench=add Testbed: Alibaba g8y instance based on Yitian 710 CPU Results: add4x4_idct_c: 698 add4x4_idct_neon: 386 add4x4_idct_sve2: 345 Command executed: ./checkasm8 --bench=zigzag Testbed: Alibaba g8y instance based on Yitian 710 CPU Results: zigzag_interleave_8x8_cavlc_frame_c: 582 zigzag_interleave_8x8_cavlc_frame_neon: 273 zigzag_interleave_8x8_cavlc_frame_sve: 257 Command executed: ./checkasm8 --bench=zigzag Testbed: AWS Graviton3 Results: zigzag_interleave_8x8_cavlc_frame_c: 587 zigzag_interleave_8x8_cavlc_frame_neon: 257 zigzag_interleave_8x8_cavlc_frame_sve: 249
This commit is contained in:
David Chen
parent
b6190c6fa1
commit
5c382660fb
8
Makefile
8
Makefile
|
|
@ -160,7 +160,7 @@ endif
|
|||
OBJCHK += tools/checkasm-arm.o
|
||||
endif
|
||||
|
||||
# AArch64 NEON optims
|
||||
# AArch64 NEON and SVE/SVE2 optims
|
||||
ifeq ($(SYS_ARCH),AARCH64)
|
||||
SRCASM_X = common/aarch64/bitstream-a.S \
|
||||
common/aarch64/cabac-a.S \
|
||||
|
|
@ -170,6 +170,12 @@ SRCASM_X = common/aarch64/bitstream-a.S \
|
|||
common/aarch64/pixel-a.S \
|
||||
common/aarch64/predict-a.S \
|
||||
common/aarch64/quant-a.S
|
||||
ifneq ($(findstring HAVE_SVE 1, $(CONFIG)),)
|
||||
SRCASM_X += common/aarch64/dct-a-sve.S
|
||||
endif
|
||||
ifneq ($(findstring HAVE_SVE2 1, $(CONFIG)),)
|
||||
SRCASM_X += common/aarch64/dct-a-sve2.S
|
||||
endif
|
||||
SRCS_X += common/aarch64/asm-offsets.c \
|
||||
common/aarch64/mc-c.c \
|
||||
common/aarch64/predict-c.c
|
||||
|
|
|
|||
|
|
@ -0,0 +1,88 @@
|
|||
/****************************************************************************
|
||||
* dct-a-sve.S: aarch64 transform and zigzag
|
||||
*****************************************************************************
|
||||
* Copyright (C) 2009-2023 x264 project
|
||||
*
|
||||
* Authors: David Chen <david.chen@myais.com.cn>
|
||||
*
|
||||
* This program is free software; you can redistribute it and/or modify
|
||||
* it under the terms of the GNU General Public License as published by
|
||||
* the Free Software Foundation; either version 2 of the License, or
|
||||
* (at your option) any later version.
|
||||
*
|
||||
* This program is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
* GNU General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with this program; if not, write to the Free Software
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
|
||||
*
|
||||
* This program is also available under a commercial proprietary license.
|
||||
* For more information, contact us at licensing@x264.com.
|
||||
*****************************************************************************/
|
||||
|
||||
#include "asm.S"
|
||||
#include "dct-a-common.S"
|
||||
|
||||
.arch armv8-a+sve
|
||||
|
||||
function sub4x4_dct_sve, export=1
|
||||
mov x3, #FENC_STRIDE
|
||||
mov x4, #FDEC_STRIDE
|
||||
ptrue p0.h, vl4
|
||||
ld1b {z0.h}, p0/z, [x1]
|
||||
add x1, x1, x3
|
||||
ld1b {z1.h}, p0/z, [x2]
|
||||
add x2, x2, x4
|
||||
ld1b {z2.h}, p0/z, [x1]
|
||||
add x1, x1, x3
|
||||
sub v16.4h, v0.4h, v1.4h
|
||||
ld1b {z3.h}, p0/z, [x2]
|
||||
add x2, x2, x4
|
||||
ld1b {z4.h}, p0/z, [x1]
|
||||
add x1, x1, x3
|
||||
sub v17.4h, v2.4h, v3.4h
|
||||
ld1b {z5.h}, p0/z, [x2]
|
||||
add x2, x2, x4
|
||||
ld1b {z6.h}, p0/z, [x1]
|
||||
sub v18.4h, v4.4h, v5.4h
|
||||
ld1b {z7.h}, p0/z, [x2]
|
||||
sub v19.4h, v6.4h, v7.4h
|
||||
|
||||
DCT_1D v0.4h, v1.4h, v2.4h, v3.4h, v16.4h, v17.4h, v18.4h, v19.4h
|
||||
transpose4x4.h v0, v1, v2, v3, v4, v5, v6, v7
|
||||
DCT_1D v4.4h, v5.4h, v6.4h, v7.4h, v0.4h, v1.4h, v2.4h, v3.4h
|
||||
st1 {v4.4h,v5.4h,v6.4h,v7.4h}, [x0]
|
||||
ret
|
||||
endfunc
|
||||
|
||||
function zigzag_interleave_8x8_cavlc_sve, export=1
|
||||
mov z31.s, #1
|
||||
ptrue p2.s, vl2
|
||||
ld4 {v0.8h,v1.8h,v2.8h,v3.8h}, [x1], #64
|
||||
ld4 {v4.8h,v5.8h,v6.8h,v7.8h}, [x1], #64
|
||||
umax v16.8h, v0.8h, v4.8h
|
||||
umax v17.8h, v1.8h, v5.8h
|
||||
umax v18.8h, v2.8h, v6.8h
|
||||
umax v19.8h, v3.8h, v7.8h
|
||||
st1 {v0.8h}, [x0], #16
|
||||
st1 {v4.8h}, [x0], #16
|
||||
umaxp v16.8h, v16.8h, v17.8h
|
||||
umaxp v18.8h, v18.8h, v19.8h
|
||||
st1 {v1.8h}, [x0], #16
|
||||
st1 {v5.8h}, [x0], #16
|
||||
umaxp v16.8h, v16.8h, v18.8h
|
||||
st1 {v2.8h}, [x0], #16
|
||||
st1 {v6.8h}, [x0], #16
|
||||
cmhs v16.4s, v16.4s, v31.4s
|
||||
st1 {v3.8h}, [x0], #16
|
||||
and v16.16b, v16.16b, v31.16b
|
||||
st1 {v7.8h}, [x0], #16
|
||||
st1b {z16.s}, p2, [x2]
|
||||
add x2, x2, #8
|
||||
mov v16.d[0], v16.d[1]
|
||||
st1b {z16.s}, p2, [x2]
|
||||
ret
|
||||
endfunc
|
||||
|
|
@ -0,0 +1,89 @@
|
|||
/****************************************************************************
|
||||
* dct-a-sve2.S: aarch64 transform and zigzag
|
||||
*****************************************************************************
|
||||
* Copyright (C) 2009-2023 x264 project
|
||||
*
|
||||
* Authors: David Chen <david.chen@myais.com.cn>
|
||||
*
|
||||
* This program is free software; you can redistribute it and/or modify
|
||||
* it under the terms of the GNU General Public License as published by
|
||||
* the Free Software Foundation; either version 2 of the License, or
|
||||
* (at your option) any later version.
|
||||
*
|
||||
* This program is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
* GNU General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with this program; if not, write to the Free Software
|
||||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
|
||||
*
|
||||
* This program is also available under a commercial proprietary license.
|
||||
* For more information, contact us at licensing@x264.com.
|
||||
*****************************************************************************/
|
||||
|
||||
#include "asm.S"
|
||||
#include "dct-a-common.S"
|
||||
|
||||
.arch armv8-a+sve+sve2
|
||||
|
||||
function add4x4_idct_sve2, export=1
|
||||
mov x2, #FDEC_STRIDE
|
||||
mov x11, x0
|
||||
ptrue p0.h, vl8
|
||||
ptrue p1.h, vl4
|
||||
ld1 {v0.8h, v1.8h}, [x1]
|
||||
|
||||
SUMSUB_AB v4.8h, v5.8h, v0.8h, v1.8h
|
||||
|
||||
sshr v7.8h, v0.8h, #1
|
||||
sshr v6.8h, v1.8h, #1
|
||||
sub v7.8h, v7.8h, v1.8h
|
||||
add v6.8h, v6.8h, v0.8h
|
||||
mov v7.d[0], v7.d[1]
|
||||
mov v6.d[0], v6.d[1]
|
||||
ld1b {z28.h}, p0/z, [x11]
|
||||
add x11, x11, x2
|
||||
SUMSUB_AB v0.8h, v2.8h, v4.8h, v6.8h
|
||||
SUMSUB_AB v1.8h, v3.8h, v5.8h, v7.8h
|
||||
|
||||
transpose4x4.h v0, v1, v3, v2, v16, v17, v18, v19
|
||||
|
||||
SUMSUB_AB v4.4h, v5.4h, v0.4h, v3.4h
|
||||
|
||||
sshr v7.4h, v1.4h, #1
|
||||
sshr v6.4h, v2.4h, #1
|
||||
sub v7.4h, v7.4h, v2.4h
|
||||
add v6.4h, v6.4h, v1.4h
|
||||
ld1b {z29.h}, p0/z, [x11]
|
||||
add x11, x11, x2
|
||||
SUMSUB_AB v0.4h, v2.4h, v4.4h, v6.4h
|
||||
SUMSUB_AB v1.4h, v3.4h, v5.4h, v7.4h
|
||||
|
||||
srshr z0.h, p1/m, z0.h, #6
|
||||
srshr z1.h, p1/m, z1.h, #6
|
||||
ld1b {z31.h}, p0/z, [x11]
|
||||
add x11, x11, x2
|
||||
srshr z2.h, p1/m, z2.h, #6
|
||||
srshr z3.h, p1/m, z3.h, #6
|
||||
ld1b {z30.h}, p0/z, [x11]
|
||||
|
||||
add v0.8h, v0.8h, v28.8h
|
||||
add v1.8h, v1.8h, v29.8h
|
||||
add v2.8h, v2.8h, v30.8h
|
||||
add v3.8h, v3.8h, v31.8h
|
||||
sqxtunb z0.b, z0.h
|
||||
sqxtunb z1.b, z1.h
|
||||
sqxtunb z2.b, z2.h
|
||||
sqxtunb z3.b, z3.h
|
||||
|
||||
st1b {z0.h}, p1, [x0]
|
||||
add x0, x0, x2
|
||||
st1b {z1.h}, p1, [x0]
|
||||
add x0, x0, x2
|
||||
st1b {z3.h}, p1, [x0]
|
||||
add x0, x0, x2
|
||||
st1b {z2.h}, p1, [x0]
|
||||
ret
|
||||
endfunc
|
||||
|
|
@ -91,4 +91,13 @@ int x264_zigzag_sub_8x8_frame_neon( dctcoef level[16], const pixel *p_src, pixel
|
|||
#define x264_zigzag_interleave_8x8_cavlc_neon x264_template(zigzag_interleave_8x8_cavlc_neon)
|
||||
void x264_zigzag_interleave_8x8_cavlc_neon( dctcoef *dst, dctcoef *src, uint8_t *nnz );
|
||||
|
||||
#define x264_sub4x4_dct_sve x264_template(sub4x4_dct_sve)
|
||||
void x264_sub4x4_dct_sve( int16_t dct[16], uint8_t *pix1, uint8_t *pix2 );
|
||||
|
||||
#define x264_add4x4_idct_sve2 x264_template(add4x4_idct_sve2)
|
||||
void x264_add4x4_idct_sve2( uint8_t *p_dst, int16_t dct[16] );
|
||||
|
||||
#define x264_zigzag_interleave_8x8_cavlc_sve x264_template(zigzag_interleave_8x8_cavlc_sve)
|
||||
void x264_zigzag_interleave_8x8_cavlc_sve( dctcoef *dst, dctcoef *src, uint8_t *nnz );
|
||||
|
||||
#endif
|
||||
|
|
|
|||
18
common/dct.c
18
common/dct.c
|
|
@ -707,6 +707,18 @@ void x264_dct_init( uint32_t cpu, x264_dct_function_t *dctf )
|
|||
dctf->add16x16_idct8= x264_add16x16_idct8_neon;
|
||||
dctf->sub8x16_dct_dc= x264_sub8x16_dct_dc_neon;
|
||||
}
|
||||
#if HAVE_SVE
|
||||
if ( cpu&X264_CPU_SVE )
|
||||
{
|
||||
dctf->sub4x4_dct = x264_sub4x4_dct_sve;
|
||||
}
|
||||
#endif
|
||||
#if HAVE_SVE2
|
||||
if ( cpu&X264_CPU_SVE2 )
|
||||
{
|
||||
dctf->add4x4_idct = x264_add4x4_idct_sve2;
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#if HAVE_MSA
|
||||
|
|
@ -1105,6 +1117,12 @@ void x264_zigzag_init( uint32_t cpu, x264_zigzag_function_t *pf_progressive, x26
|
|||
pf_interlaced->interleave_8x8_cavlc =
|
||||
pf_progressive->interleave_8x8_cavlc = x264_zigzag_interleave_8x8_cavlc_neon;
|
||||
}
|
||||
#if HAVE_SVE
|
||||
if( cpu&X264_CPU_SVE )
|
||||
{
|
||||
pf_progressive->interleave_8x8_cavlc = x264_zigzag_interleave_8x8_cavlc_sve;
|
||||
}
|
||||
#endif
|
||||
#endif // HAVE_AARCH64
|
||||
|
||||
#if HAVE_ALTIVEC
|
||||
|
|
|
|||
Loading…
Reference in New Issue